test server with invalid port number

[icing]

I did this

make a PR, have CI jobs run. Sometimes the tests fail with the following in the log:

[0-x] == Info: Unsupported proxy syntax in '127.0.0.1:[not': Port number was not a decimal number between 0 and 65535

Sporadically. There seems to be a race somewhere in retrieving ports numbers from started servers.

I expected the following

reliable port recoognition or report that a test server did not start properly.

curl/libcurl version

curl 8.14.1.-DEV

operating system

github CI

[bagder]

The http-proxy server is supposed to be started completely synchronously before the test case runs, and once started, the port should be known...

curl/tests/servers.pm

Lines 2452 to 2470 in 54834b4

	elsif($what eq "http-proxy") {
	if($run{'http-proxy'} &&
	!responsive_http_server("http", $verbose, "proxy",
	protoport("httpproxy"))) {
	if(stopserver('http-proxy')) {
	return ("failed stopping unresponsive HTTP-proxy server", 3);
	}
	}
	if(!$run{'http-proxy'}) {
	($serr, $pid, $pid2, $PORT{"httpproxy"}) =
	runhttpserver("http", $verbose, "proxy");
	if($pid <= 0) {
	return ("failed starting HTTP-proxy server", $serr);
	}
	logmsg sprintf ("* pid http-proxy => %d %d\n", $pid, $pid2)
	if($verbose);
	$run{'http-proxy'}="$pid $pid2";
	}
	}

[icing]

If I add a sleep(1000); in util.c:349 I can reproduce this.

The server has started, but the writing of the port file takes more time. The runner needs to wait for the portfile to give a meaningful result or fail the start. I try to add something.

[icing]

#17495 should fix this.